# Joseph Kim

Markdown copy of https://realalpha.app/people/joseph-kim/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

Real Alpha has cut 8 moments of Joseph Kim (Head of Financial Services, OpenAI) talking, from an interview on Latent Space published Oct 31, 2026.

- Role: Head of Financial Services, OpenAI

## What Joseph Kim said

- [Bundled market data fixed connector failures](https://realalpha.app/s/bundled-market-data-fixed-connector-failures/)
  Latent Space · Oct 31, 2026 · 3 min
  > “Some of the MCPs that we connected, some big names, 40% plus error rate on their MCPs, right?” — Joseph Kim · [6:03](https://realalpha.app/v/v-d1WTqIULs/?t=363)
- [Most finance AI product work is buildable by startups](https://realalpha.app/s/most-finance-ai-product-work-is-buildable-by-startups/)
  Latent Space · Oct 31, 2026 · 2 min
  > “95% of what we built was something that absolutely a startup could do, another company could do.” — Joseph Kim · [7:16](https://realalpha.app/v/v-d1WTqIULs/?t=436)
- [Financial AI evals must predict real job performance](https://realalpha.app/s/financial-ai-evals-must-predict-real-job-performance/)
  Latent Space · Oct 31, 2026 · 3 min
  > “building a good eval is incredibly hard. That's not obvious to people who haven't worked on an eval before.” — Joseph Kim · [22:56](https://realalpha.app/v/v-d1WTqIULs/?t=1376)
- [Financial AI can fit into existing review hierarchies](https://realalpha.app/s/financial-ai-can-fit-into-existing-review-hierarchies/)
  Latent Space · Oct 31, 2026 · 3 min
  > “So if models behave more like people, then you can leverage those systems.” — Joseph Kim · [11:46](https://realalpha.app/v/v-d1WTqIULs/?t=706)
- [Moving up the analyst stack makes evaluation harder](https://realalpha.app/s/moving-up-the-analyst-stack-makes-evaluation-harder/)
  Latent Space · Oct 31, 2026 · 3 min
  > “We really want to get the P99.9 type of feedback to make the model as good as humans.” — Joseph Kim · [21:41](https://realalpha.app/v/v-d1WTqIULs/?t=1301)
- [Pairwise human feedback can train models on subjective taste](https://realalpha.app/s/pairwise-human-feedback-can-train-models-on-subjective-taste/)
  Latent Space · Oct 31, 2026 · 3 min
  > “You get a bunch of humans to look at pairs of things and say which one is better.” — Joseph Kim · [16:31](https://realalpha.app/v/v-d1WTqIULs/?t=991)
- [Factual errors are more dangerous than weak visual judgment](https://realalpha.app/s/factual-errors-are-more-dangerous-than-weak-visual-judgment/)
  Latent Space · Oct 31, 2026 · 3 min
  > “the career-ending mistake is something where you get a piece of information factually incorrect. And that piece of factual information changes the conclusion.” — Joseph Kim · [13:59](https://realalpha.app/v/v-d1WTqIULs/?t=839)
- [Finance training improves the shared model, not just one product](https://realalpha.app/s/finance-training-improves-the-shared-model-not-just-one/)
  Latent Space · Oct 31, 2026 · 2 min
  > “all of the training data that we do for financial reasoning analysis, we do upstream into what we call the big run” — Joseph Kim · [2:14](https://realalpha.app/v/v-d1WTqIULs/?t=134)

## The episode

- [Building ChatGPT for Finance — Joseph Kim, OpenAI's Head of Financial Services](https://realalpha.app/v/v-d1WTqIULs/) · Latent Space · Oct 31, 2026